[Bug]: probe failed: HuggingFaceEngine only supports HuggingFace backend

[bogdansolga]

Shimmy Version

0.1.0

Installation Method

Downloaded binary from GitHub

What happened?

shimmy list
📋 Registered Models:
phi3-lora => "/Volumes/NVMe/Development/LLMs/Phi-3-mini-4k-instruct-fp16.gguf"

🔍 Auto-Discovered Models:
phi-3-mini-4k-instruct-q4 => "./Phi-3-mini-4k-instruct-q4.gguf" [2282MB]
phi-3-mini-4k-instruct-fp16 => "./Phi-3-mini-4k-instruct-fp16.gguf" [7289MB]

---

shimmy probe phi-3-mini-4k-instruct-fp16
probe failed: HuggingFaceEngine only supports HuggingFace backend

---

shimmy probe phi3-lora
probe failed: HuggingFaceEngine only supports HuggingFace backend

Steps to Reproduce

1. Download Shimmy as release binary
2. Download Phi3, as described in the readme: huggingface-cli download microsoft/Phi-3-mini-4k-instruct-gguf --local-dir .
3. Try to use it

result: probe failed: HuggingFaceEngine only supports HuggingFace backend

Relevant logs

Operating System

macOS (Apple Silicon)

Code of Conduct

• I agree to follow this project's Code of Conduct

[Michael-A-Kuykendall]

✅ RESOLVED in v0.1.1

This issue has been fixed! The problem was in the backend selection logic - .gguf files were being incorrectly routed to the HuggingFace engine instead of the Llama engine.

What was fixed:

• Modified InferenceEngineAdapter::select_backend() to properly detect .gguf files by extension
• All .gguf files now correctly route to the Llama engine instead of failing with HuggingFace

How to get the fix:

# Update to the latest version
cargo install shimmy --force

# Or build from source
git pull origin main
cargo build --release

Verification:

shimmy probe phi-3-mini-4k-instruct-fp16  # Should work now! ✅

The fix is included in shimmy v0.1.1 published to crates.io and the latest main branch.

[bogdansolga]

ETA for binary files, for the release 0.1.1?

[Michael-A-Kuykendall]

Complete!

…

On Tue, Sep 9, 2025 at 10:20 AM Bogdan Solga ***@***.***> wrote: *bogdansolga* left a comment (Michael-A-Kuykendall/shimmy#6) <#6 (comment)> ETA for binary files, for the release 0.1.1? — Reply to this email directly, view it on GitHub <#6 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/BO4RGDPUXF2O2P4RSEU3WMD3R3V2RAVCNFSM6AAAAACF76FYGOVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZTENZRGIYDENJWGA> . You are receiving this because you modified the open/close state.Message ID: ***@***.***>

[bogdansolga]

I don't understand

[Michael-A-Kuykendall]

This has been resolved, I am not doing binaries for old versions that lacked them... you can get the newest here;

https://github.com/Michael-A-Kuykendall/shimmy/releases/tag/v1.0.0

[bogdansolga]

This has been resolved, I am not doing binaries for old versions that lacked them... you can get the newest here;

https://github.com/Michael-A-Kuykendall/shimmy/releases/tag/v1.0.0

These are the artifacts for which I have reported the issue.

You said you fixed the issue and you published the source code for them. I am not asking for binaries of the old versions, I am asking for binaries of the source code that you have published 2 hours ago 🙂

[Michael-A-Kuykendall]

I had some publishing issues for earlier versions, are you asking me to roll back and produce a binary for an older version? Now I am confused.

[Michael-A-Kuykendall]

[bogdansolga]

Timeline:

1. September 5th:

• action: wanted to download the MacOS binaries, they were not available
• follow-up: created MacOS and Linux distributions #1

2. September 8th:

• action: you have posted the release https://github.com/Michael-A-Kuykendall/shimmy/releases/tag/v1.0.0

3. September 9th:

• action: downloaded the ARM MacOS binary and tried it --> didn't work, HuggingFaceEngine only supports HuggingFace backend
  • although you named it v1.0.0, shimmy --version returns shimmy 0.1.0
• follow-up: created [Bug]: probe failed: HuggingFaceEngine only supports HuggingFace backend #6

4. September 9th:

• action: you closed [Bug]: probe failed: HuggingFaceEngine only supports HuggingFace backend #6, stating that the issue is RESOLVED in v0.1.1 and that The fix is included in shimmy v0.1.1 published to crates.io and the latest main branch.

• follow-up: I asked about the binaries, as the release https://github.com/Michael-A-Kuykendall/shimmy/releases/tag/v1.1.0 contains just the source code

  • you stated This has been resolved, I am not doing binaries for old versions that lacked them... you can get the newest here - https://github.com/Michael-A-Kuykendall/shimmy/releases/tag/v1.0.0
  • I explained that These are the artifacts for which I have reported the issue, and explained that I have asked about the new binaries, not the ones that already exist, and are returning shimmy 0.1.0 as version

The root cause of these misunderstandings may be that the binaries from the release https://github.com/Michael-A-Kuykendall/shimmy/releases/tag/v1.0.0 are returning shimmy 0.1.0 as version, not 1.0.0, as you may have intended and/or assumed that they return.

Please let me know if these details are clarifying the confusion(s)